SMED: mastering quick changeover on the shop floor
Your customer wants ever-shorter runs. Every changeover is a stopped machine. SMED lives exactly in that collision.
SMED (single-minute exchange of dies, or quick changeover) is the lean method for cutting the time a machine sits stopped between the last part of one batch and the first good part of the next. Shigeo Shingo developed it at Toyota, and the goal names the technique: changeovers in under ten minutes — a single digit.
Sebastian Brau's blog
Enter your email and you will get an alert when something new is published. Unsubscribe anytime.
Why should you care? Because the market keeps asking you for shorter runs and changing part numbers, and every changeover is a stopped machine. If changing over costs two hours, you'll run big batches to spread that cost — and with big batches come the inventory, the waiting and the rest of the bill.
The idea that changes everything: internal versus external
The heart of SMED is a distinction that looks stupid, it's so simple:
| Type of operation | When it's done | Examples |
|---|---|---|
| Internal | with the machine stopped | release the die, mount the new one, adjust |
| External | with the machine running | fetch the die, stage the tools, preheat |
In an unexamined changeover, almost everything happens with the machine stopped — including the part that doesn't need it. The operator stops the machine and then goes looking for the fixture, the wrench and the cart. The SMED sequence attacks that in order:
- Film a real changeover and break it into operations. Without filming it doesn't work: the memory of a changeover is always shorter than the changeover.
- Convert to external everything that can be done with the machine running. This alone tends to cut times in half.
- Simplify what stays internal: quick clamps instead of bolts, adjustments eliminated with fixed stops, two people in parallel where one used to go.
- Standardize and measure. The good changeover gets documented, and the time gets logged on every changeover — not at the quarterly event.
The catch?
SMED fails for the same two reasons as always, and neither is technical.
First: rushing in without preparation. A quick-changeover programme thrown together without a team — the tooling technician, the line operator, maintenance — dies within three months. Second: not standardizing. Somebody hits a twenty-minute changeover on one glorious Tuesday, nobody documents it, and six weeks later you're back to the usual hour and a half.
And a condition you already know from the other tools in this house: if management doesn't treat it as a priority, nobody else will.
The bridge to 2026
SMED is from 1969 and gains importance every year, for a reason Shingo couldn't entirely foresee: variety is no longer negotiable. Short runs, customization, part numbers coming and going. The factory that changes over fast can say yes; the one that can't builds to stock and prays.
And there's a symmetry worth pointing out. A changeover is to a machine what retraining is to an industrial AI system: the time it isn't producing because you're adapting it to something else. That's why I argue for AI systems that train in weeks and re-adapt when the product changes — it's SMED applied to the model. A flexible factory with a rigid AI is tied down again, only now the rope is made of software.